Section 2 - Similarity Search

---

Section 2 describes routines within Polymetrics used for similarity searches. Similarity search can be used for

1. Application/Product development - identifying alternatives based on specified criteria.
2. Data Selection - selecting balanced datasets for supervised learning.
3. Competitive Analysis - Reviewing competitive offerings and identifying white spaces.

Presently similarity search within Polymetrics uses the kNN method with a predefined tolerance for identifying similar products. The data is visualized in an interactive heatmap and can also be obtained in a tabulated form.

In addition to Polymetrics, the following libraries are imported:

• numpy and pandas - for data handling and analysis.
• scikit-learn - for the nearest neighbour procedure.
• plotly - for plotting interactive heatmaps.
• IPython - for creating tabulated outputs of DataFrames.

import pandas as pd
import Polymetrics as poly
import FileImport
import polymetrics_config
from IPython.display import display

2.1 Importing Polymetrics Dataset

---

The data used in the example below is compiled using the Material Search option in UL Prospector (https://www.ulprospector.com/).
A few PE resins from the following manufacturers were included in the analysis.

1. Chevron Phillips Chemical Company, LLC (https://www.cpchem.com/)
2. Dow Inc, (https://www.dow.com)
3. SABIC (https://www.sabic.com/en)
4. ExxonMobil Chemical Company (https://www.exxonmobilchemical.com/)

The data was added to the excel file - 'Example_Dataset.xlsx' with the project name 'kNN', and is imported using the FileImport module in Polymetrics.

The XLSXImport function imports the data and conditions some datatypes for easier processing. A short preview of the dataset is shown below.

df_in = FileImport.XLSXImport("Article/Example_Dataset.xlsx", sheet_name = 'Data')
result = df_in[(df_in['Project'] == 'kNN')]
df_kNN = result.dropna(axis = 1)
display(df_kNN.tail(5))

	Identifier	Name	UID	Project	Type	Density	Tm	I2
75	DOWLEX_GM_8091	DOWLEX™ GM 8091	687A	kNN	Resin_Commercial	0.918	111.0	1.0
76	DOWLEX_NG_5045P	DOWLEX™ NG 5045P	6DYZ	kNN	Resin_Commercial	0.917	118.0	0.8
77	ELITE_AT_6111	ELITE™ AT 6111	6EXO	kNN	Resin_Commercial	0.912	109.0	3.7
78	Marlex_5428	Marlex® 5428	6K0I	kNN	Resin_Commercial	0.930	111.0	2.2
79	Marlex_5430	Marlex® 5430	6LYJ	kNN	Resin_Commercial	0.925	111.0	2.2

2.2 Function: Similarity Matrix

---

simiarity_matrix function takes DataFrame as an input, scales the data, and uses the kNN method to calculate distances between samples. A cut-off is determined by selecting a certain quantile of the pairwise distance data as a threshold. For example, Q = 0.25 would suggest that the cut-off value is the 1st quartile of the pairwise distance data. The pairs with distances lower than the threshold value are selected for further analysis.

In the example below, the similarity between PE resins is determined by density, MI, and peak melting temperatures as reported in UL Prospector. Only the first quartile of the data is considered for further evaluation. The result is plotted in the form of an interactive heatmap.

indices, distances =  poly.similarity_matrix(df_kNN, label = 'UID', Q = 0.25, plot = True)
#Function signature 
# label = label on the heatmap.
# Q = quantile value.
# plot = boolean, turn on/off heatmap.  

2.2 Function: Find Similar

---

find_similar function uses the outputs of similarity_matrix and returns a list of polymers similar to the query polymer arranged by distances in ascending order.

display(poly.find_similar('3A5L', df_kNN, indices, distances, label = 'UID'))

	Identifier	Name	UID	Project	Type	Density	Tm	I2	Distances
41	Marlex_5626	Marlex® 5626	3A5L	kNN	Resin_Commercial	0.922	114.0	0.65	0.000000
74	Marlex_5754	Marlex® 5754	5YIM	kNN	Resin_Commercial	0.925	112.0	0.80	0.079563
31	SUPEER_8118	SUPEER™ 8118	2PW6	kNN	Resin_Commercial	0.918	115.0	1.10	0.109134
32	SUPEER_8118L	SUPEER™ 8118L	2R3Z	kNN	Resin_Commercial	0.918	115.0	1.10	0.109134
71	DOWLEX_2688G	DOWLEX™ 2688G	5PN2	kNN	Resin_Commercial	0.917	117.0	0.50	0.120966
75	DOWLEX_GM_8091	DOWLEX™ GM 8091	687A	kNN	Resin_Commercial	0.918	111.0	1.00	0.124252
73	DOWLEX_GM_8071G	DOWLEX™ GM 8071G	5WDK	kNN	Resin_Commercial	0.920	118.0	0.90	0.124591
48	ELITE_AT_6501	ELITE™ AT 6501	3I4Y	kNN	Resin_Commercial	0.914	115.0	0.85	0.140967
76	DOWLEX_NG_5045P	DOWLEX™ NG 5045P	6DYZ	kNN	Resin_Commercial	0.917	118.0	0.80	0.141541
55	DOWLEX_2045G_CIR	DOWLEX™ 2045G CIR	42XQ	kNN	Resin_Commercial	0.920	119.0	1.00	0.156368
56	DOWLEX_2045LC	DOWLEX™ 2045LC	4CD2	kNN	Resin_Commercial	0.920	119.0	1.00	0.156368
57	DOWLEX_2645.11S	DOWLEX™ 2645.11S	4DIP	kNN	Resin_Commercial	0.921	120.0	0.90	0.173556
63	DOWLEX_HMS_8017	DOWLEX™ HMS 8017	4SQL	kNN	Resin_Commercial	0.918	121.0	0.75	0.206358
53	DOWLEX_2256G	DOWLEX™ 2256G	3U15	kNN	Resin_Commercial	0.920	121.0	1.00	0.207290
67	DOWLEX_GM_8070G	DOWLEX™ GM 8070G	572N	kNN	Resin_Commercial	0.917	121.0	0.90	0.216377
52	DOWLEX_2045.11S	DOWLEX™ 2045.11S	3NX3	kNN	Resin_Commercial	0.922	122.0	1.00	0.231154
51	DOWLEX_2045.11G	DOWLEX™ 2045.11G	3NK6	kNN	Resin_Commercial	0.922	122.0	1.00	0.231154
46	DOWLEX_2045	DOWLEX™ 2045	3EPY	kNN	Resin_Commercial	0.920	122.0	1.00	0.233545
50	ELITE_AT_6410	ELITE™ AT 6410	3MSN	kNN	Resin_Commercial	0.912	108.0	0.85	0.238491
49	DOWLEX_2601A	DOWLEX™ 2601A	3KE1	kNN	Resin_Commercial	0.924	122.0	1.30	0.253891
47	DOWLEX_2070G	DOWLEX™ 2070G	3HBQ	kNN	Resin_Commercial	0.922	123.0	1.00	0.257972
62	ELITE_5110G	ELITE™ 5110G	4S5N	kNN	Resin_Commercial	0.926	123.0	0.85	0.261279